AI Hacker — XP.COM, LLC

Own the engine.
Then ship the products.

Most AI companies rent compute by the image and build one product on top of it. We bought the silicon, wrote the engine, and run a portfolio against it. Adding a product costs a database row and a landing page, not a codebase, a deployment, or a new bill. That is not a plan. It is how eighteen live branded sites already run today.

Brains.biz, from nothing to selling

Our newest product is the cleanest test of the model. It needed no new infrastructure, because the accounts, credits, storage, queue and render fleet already existed. What it needed was a landing page, one pipeline, and the vertical detail that makes the output publishable.

Days

From no product to a live storefront with finished example books

Book types shipping: how-to, novel, picture book, coloring book

~30 min

Researched, written, illustrated, and exported to three KDP files

cents

Marginal cost of a finished illustrated book on owned hardware

The same spine is what makes the next product cheap. That is the entire argument, and brains.biz is the receipt.

What is running today

Status is stated as it is. A product with a finished storefront and no customers says so, because you will find out anyway.

The Bell Gardener, a towering wooden and brass Titan with a bell-shaped hat, stilt legs, and six playable machinery rooms inside its body. — Bell Gardener: actual level render from the experimental Android build.

Featured game

Rope Ninja

Two ropes. A tiny explorer. Giants that need your help.

Swing through the bodies of enormous mechanical Titans. Tie parts together, free jammed gears, carry supplies, and help the workers who keep these giants running. Repairs bring the machinery back to life. Corrupted Titans need to be stopped first.

A playable puzzle adventure with an in-game level editor and a growing Titan authoring pipeline. Asset Yard supplies generated art, voices, music, and sound effects; device playtests guide the design.

In development · Android testing

Explore Rope Ninja Publisher enquiries

Product	What it does	Status
Rope Ninjaropeninja.com	A two-rope puzzle adventure inside giant mechanical Titans. Playable Android builds, an in-game level editor, and an experimental pipeline for creating new Titan encounters.	In development
UNCEN.AIuncen.ai	An AI video studio. One written idea becomes a finished multi-scene production, with story, images, voices, music and the final edit generated and assembled by the platform. It is also the engine every other product runs on.	Live
Brains.bizbrains.biz	Researches what Amazon shoppers search for, writes and illustrates the book, and returns the three files Kindle Direct Publishing asks for: a Kindle EPUB, a print-ready interior PDF, and a store cover. Our newest working product.	Live
Asset Yardassetyard.dev	A command line asset service for game developers. A text prompt returns a textured, rigged, animated 3D model, plus images, voices, sound effects, music and video, straight into the project folder. Shares one account and one credit balance with the platform.	Launch preview
Swiftest.AIswiftest.ai	Answers one question through ten separate cognitive lenses, from memory and code to simulation and critique, then synthesises a single answer.	Live
SF Gearsfgear.com	A listening channel for complete science fiction audio series: fixed episode counts, recurring characters, an original score, and an actual ending. Storefront finished, catalogue in production.	Pre-launch
Multiplayers.AImultiplayers.ai	Browse browser games other people made, remix one into your own, publish it to its own subdomain. Builder and level editor complete.	Pre-launch
Twelve moreand 216 domains held	Pet studios, a chat-first assistant, a DOS game revival, a crypto news desk, portfolio sites and legal tooling, each already routed and serving from the same application.	Various

What has actually been built

Measured from the production database, September 2026.

Generation

5,849 pipeline runs at a 92% completion rate, across 6,504 projects and 236 people who have run one. Fourteen distinct pipelines in production use, from music video to 3D asset to KDP book.

Owned compute

Four heartbeating hosts carrying 104 GB of GPU memory, scheduled by our own dispatcher across 84 enabled worker types. We buy silicon, not per-image API calls, which is what makes a finished book cost cents.

Model independence

Every worker's language model is a database row: 43 active routes over 6 models and 2 providers today, with 25 generation backends wired. Swapping a model is configuration, so a price cut or a better model is captured the same week.

Character consistency

87,577 story-bible entries. A character is drawn once and that drawing is attached to every later image, so a cast stays identical across a film, a series, or a picture book. This is the hard part of illustrated generative work, and it is finished.

Print engineering

Six trim sizes, a binding margin that steps with page count, a two-pass layout resolving the circular dependency between margin and page count, and a read-back proof that every font is embedded. The desktop apps authors buy for this cost $147 and $250, and write nothing.

Output to date

254 hours of generated music and 77 hours of speech, alongside images, video and 3D. The large majority is our own research and testing rather than customer demand, and we would rather say so than imply otherwise.

A new product costs a database row

One application, many brands

Brands are routed by the incoming hostname against database rows. Eighteen sites, ten routing behaviours, one deployment. Standing up the KDP product end to end took days rather than a quarter, because the accounts, credits, storage, queue and render fleet already existed. Every product we add makes the next one cheaper.

We own the expensive part

Competitors meter what they buy per image. We meter what we already own. A finished illustrated book costs us cents in marginal spend, which is what lets an entry plan sell five of them for $19 and still clear a healthy margin. That gap widens with utilisation.

Vertical depth, not a wrapper

Anyone can call a model. The defensible work is the last mile: print geometry Amazon accepts, a cast that stays consistent across fifty images, a dispatcher that keeps owned GPUs busy. That is where our code is, and it does not arrive in a competitor's next model upgrade.

A portfolio hedges the picks

We do not need to guess which consumer AI category wins. We hold 216 domains and can stand a real product on the platform quickly, then put spend behind whichever finds demand. The shared engine means a product that fails costs a landing page, not a company.

The honest position

Pre-revenue

There are no active paid subscriptions. Plans are published and priced, and payment collection is the current work. Everything built so far was built without revenue.

Early on demand

Monthly active creators peaked at 308 at launch in May 2026 and are now in the tens. We have been building capability rather than buying attention, and no product has had a real marketing push.

What that buys

The expensive, slow, unglamorous half is done: the fleet, the dispatcher, the multi-brand plumbing, the vertical depth. What remains is distribution and billing, which is fast work by comparison.

The next proof

Brains.biz is the test: a narrow, paid, self-serve product in a market that already spends money on worse tools. It went from nothing to a live storefront with finished example books in days, on this platform.

Figures are read from the production database and stated as of September 2026. Generation volume is dominated by our own research and testing, and is reported as capability rather than as customer demand.

Own the engine.Then ship the products.